<p>Human action recognition is a critical task in computer vision, enabling systems to interpret and respond to human movements. This paper introduces a novel approach that utilizes Augmentations of Forman-Ricci Curvature (AFRC) within Graph Neural Networks (GNNs) to enhance the recognition of actions based on skeletal data. Traditional methods often struggle with issues like information over-squashing, which can limit their effectiveness. The proposed framework addresses these challenges by improving the expressiveness of graph representations of human skeletons, allowing for more accurate action classification. Through extensive experiments on the Chalearn benchmark dataset, we demonstrate that the proposed method significantly outperforms existing techniques, achieving competitive accuracy levels while maintaining computational efficiency. The integration of AFRC not only mitigates bottleneck effects but also captures complex relationships between skeletal joints, which are essential for precise gesture recognition. The implications of this research extend to various applications, including human-computer interaction, sports analysis, and surveillance, highlighting the potential for real-time action recognition systems. Future work will focus on refining the AFRC algorithm and exploring its applicability across different domains.</p>